Meta debuts new AI silicon to power platform recommendations

Meta Platforms has unveiled a roadmap for four new custom AI chips under its MTIA (Meta Training and Inference Accelerator) program, aimed at enhancing AI capabilities across its platforms like Facebook and Instagram. The first chip, MTIA 300, is already deployed to power content ranking and recommendation algorithms. The subsequent chips—MTIA 400, 450, and 500—are designed to handle AI inference workloads, with improvements in memory capacity and low-precision AI processing. Meta plans to release these chips roughly every six months over the next two years, reflecting its rapid expansion of AI infrastructure and data center footprint. Built on the open-source RISC-V architecture with design support from Broadcom and fabricated by TSMC, these chips enable Meta to optimize AI workloads more efficiently than general-purpose processors. Starting with the MTIA 400, Meta is also designing entire computing systems around the chips, including advanced cooling solutions, to support large-scale AI operations. This iterative development approach allows Meta to quickly adapt